【作者】 陈烈；

【导师】 齐维贵；

【作者基本信息】 哈尔滨工业大学 ， 电力电子与电力传动， 2009， 博士

【摘要】 我国三北地区采暖能耗占全社会能耗的27.2%,供热不但能耗大,而且效率低,单位面积采暖能耗是发达国家的3~4倍,为落实国家节能政策,研究供热节能监控系统意义重大。本文以“十一五”国家科技支撑计划重大项目“建筑节能关键技术研究与示范”(项目编号:2006BAJ01A04)和黑龙江省科技攻关项目“基于LonWorks技术和预测控制的供热FCS研究”(项目编号:GC04A104)为课题背景,研究供热节能控制策略和开发供热节能监控装置。研究中面临的问题是:必须以节能为目标,但同时也要保证优质供热;供热过程是一复杂的动力学对象;在采用先进技术的同时,要降低产品的成本。精确的数学模型是供热系统分析与控制的重要依据。将供热过程动态分解为确定部分和随机部分,分别建立这两部分模型:通过机理分析与实验建模法相结合求取确定部分的模型,用ARIMA模型拟合供热过程模型的随机部分。供热过程建模为供热质调量调解耦和预测控制研究的基础。供热负荷预报为供热节能提供依据。针对供热负荷具有的非平稳、非线性、时滞等特性,应用时间序列、最大熵、RBF神经网络多种方法对供热负荷进行预报。其中用时间序列法对平稳化处理的随机序列进行预报,用最大熵法对非平稳随机序列进行预报,用神经网络法对非线性负荷序列进行预报。为了进一步提高预报精度,将交叉预报思想引入供热负荷预报中,即用纵向预报跟踪用户对负荷的需求,用横向预报跟踪天气的变化,然后对纵向和横向预报结果进行加权交叉。通过仿真分析、比较各方法的性能。供热系统质调、量调通道间存在耦合。给出耦合程度判定方法,对供热耦合模型进行稳态与动态耦合程度分析。分别采用传统解耦方法和时滞递归RBF神经网络解耦方法,将耦合系统解耦成两个相互无干扰的单入单出系统。其中神经网络解耦法采用改进的假近邻法预估神经网络的输入维数,解决时间序列维数确定困难的问题,通过仿真验证其静态和动态性能。鉴于预测控制算法能适应供热过程非线性、时变、时滞、不确定等特性,采用预测控制对供热过程质调通道进行控制。对传统预测控制方法DMC和GPC算法进行了改进:用DMC模型简化和预报误差校正结合的方法减少计算量,提高实时性,并解决模型失配问题;对隐式自适应广义预测控制研究,给出改进的辨识和控制算法,以满足实时性要求。进行神经网络预测控制方法的研究,设计基于神经网络的预测控制器,给出其偏差控制算法和控制律求解算更多还原

【Abstract】 In north areas of China, heating energy consumption has achieved 27.2 percent of total national energy consumption. In fact, energy consumption is not only huge, but also inefficient. Nowadays, the heating energy consumption per unit area is 2 to 3 times than developed countries. In order to implement national policies of energy-saving, it is very important to study the energy-saving monitoring system. Based on above reasons, the heating energy-saving control strategy and corresponding monitoring system are studied in this dissertation. Two projects are taken as background, which are“National Eleventh Five-year Plan Key Project of Ministry of Science and Technology in 2006: Research of Key Technology for Building Energy-saving and Its Engineering Demonstration”(Grant No. 2006BAJ01A04) and“Heilongjiang Province Research Project in 2005: Research of Heat Supply Field Control System Based on LonWorks Technology and Predictive Control Theory”(Grant No. GC04A104). There are some problems to be considered: ensuring the high-quality heating while taking energy-saving as target; Heat supply system is a complex dynamics process; Reducing the product costs when applying advanced technology.As accurate mathematical models are foundations of heat supply system analysis and control, the heat supply process is dynamically divided into the certain part and the random part, which models are established separately. The model of the certain part is established by combining mechanism analysis and experimental methods, while the random part of heat supply process model is fitted with ARMA model. The heat supply process modeling is the premise of heat supply decoupling and predictive control.Heat load forecasting provides the basis for heating energy-saving. According to the characteristics of nonstationary, nonlinear, time-varying of heat load, several forecasting methods are proposed in this dissertation, which based on time series method, maximum entropy method and neural network theory. The random series stabilized is forecasted by time series method, the random series unstabilized is forecasted with the maximum entropy method, and the nonlinear characteristic is matched with neural network method. To improve the accuracy, crossover forecasting theory is introduced into heat load forecasting. With this method, the household demands are tracked by vertical forecasting, and the outdoor temperature is tracked by the horizontal forecasting. Finally, the performances of above algorithms are analyzed by comparing the simulation results. The coupling effect exists between quality-adjust and quantity-adjust channels. Firstly, judgment methods of coupling degree are given to analysis the static and dynamic coupling. Then, two non-interference SISO independent systems are obtained by using conventional decoupling method and decoupling method based on recurrent RBF neural network with delays respectively. The input dimensions of neural network are estimated by modified false neighborhood method. In addition, the static and dynamic performances of the methods are validated by simulation.As predictive control can meet the heating characteristics of nonlinear, changeful, time-delay, uncertain, the predictive control is applied in heat supply. The basic DMC and GPC algorithms are improved: improved DMC algorithm combined with model simplified and predictive error correction algorithm is applied to decrease computational complexity, and solve the model mismatch problem; Besides this, control algorithm of implicit adaptive GPC with improved identification algorithm is given to improve real-time ability. Further more, an intelligent predictive control method based on neural network is studied, and its deviation control algorithm along with control rate solving algorithm are presented.Finally the engineering application is studied. Through software and hardware design, scheme of heat supply monitoring system based on GPRS and PLC is given. The stability and reliability of control system is guaranteed by advanced technology and components selection. The device prototype achieved the design indexes and passed the technical inspection of national quality inspection department by tested in a substation.更多还原